Distribution plays · № 1 of 7 · borrows the attention inside AI assistants
Turn an MCP server into a sales channel
Model Context Protocol servers plug your product directly into Claude, ChatGPT and Cursor. Your app no longer only lives on your website — it can also live in the places people go to connect tools to their AI assistants.
What the evidence says
There’s a named 5-factor pattern behind successful servers: a daily pain point, a simple interface over sophisticated internals, zero-setup install, works in every MCP client, privacy by design. Focus wins — Context7 exposes exactly two tools. Hands-on Architectscase study
The catch
Know what your tool is bad at: Context7 scores 9.4/10 on single-library queries but 3.5/10 cross-library. And the protocol is young — “the worst documented technology I have ever encountered,” per one practitioner; OAuth and transport choices have real footguns. How to MCPguide
The cheap test
Build around one question users already ask their AI — a single tool, deployed remotely (hosted URL, not local install), listed on the registries. “Every additional setup step is a point where potential users drop off.”
First steps, from the essay
- Identify one problem users would rather ask an AI than click through your interface
- Build the MCP server around that single job — one useful function beats recreating the product
- Deploy remotely (hosted at a URL, not a local install)
- Publish to the MCP registry, Smithery, mcp.so, and PulseMCP
The cast
- 59k+ GitHub stars, 240k+ weekly downloads — zero traditional marketing
- Exposes exactly two tools; focus as a design principle
- Free scraping inside Claude/Cursor funnels users to the paid API
- Also recurs in play 4 as the dataset-building tool
- Publishing to all four is the play’s whole “go to market”
Go deeper
- Context7 — the living prooffirst-party
- How to MCP — the build guidecase study
- What makes an MCP server successful?case study